Merge pull request #93 from Azure-Samples/gk/classification-expenses

.gitignore

@@ -372,4 +372,8 @@ tfplan
 *.pem
 *.crt
 *.pub
-id_rsa
+id_rsa
+.vscode/
+python/expense-classification-guidance/.env
+
+python/expense-classification-guidance/.idea/

python/classify-expenses/main.py

@@ -6,7 +6,7 @@
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import RedirectResponse
 
-from routes.main import router as reddit_router
+from routes.main import router
 from settings import Settings
 
 logging.getLogger("aiohttp").setLevel(logging.ERROR)
@@ -23,7 +23,7 @@
     allow_headers=["*"],
 )
 
-app.include_router(reddit_router)
+app.include_router(router)
 
 
 @app.get("/health")

python/classify-expenses/orchestration/classify_expense.py

@@ -1,13 +1,14 @@
 from data.expense_data import ExpenseInput
 from data.orchestrator import Orchestrator
+from orchestration.guidance import guidance_classify
 
 
 async def classify_expense(expense_input: ExpenseInput):
     orchestrator = expense_input.orchestrator
     # Here you need to implement the logic for each of the orchestrators
     if orchestrator == Orchestrator.GUIDANCE:
         # classify using guidance orchestrator
-        pass
+        return guidance_classify(expense_input)
     elif orchestrator == Orchestrator.SEMANTICKERNEL:
         # classify using semantickernel orchestrator
         pass

python/classify-expenses/orchestration/guidance.py

@@ -5,16 +5,24 @@
 from data.expense_data import ExpenseInput
 from settings import settings
 
-classify_llm = guidance.llms.AzureOpenAI(
-    model="gpt-4-0314",
-    client_id=settings.openai_api_key,
-    endpoint=settings.openai_api_base)
 
-async def guidance_classify(expense_input: ExpenseInput):
+guidance.llm = guidance.llms.OpenAI(
+    'text-davinci-003',
+    api_type='azure',
+    api_key=settings.openai_api_key,
+    api_base=settings.openai_api_base,
+    api_version='2023-05-15',
+    deployment_id='gk-davinci-003',
+    caching=True
+)
+
+
+def guidance_classify(expense_input: ExpenseInput):
     # pre-define valid expense categories
     valid_categories = ["Learning", "Housing", "Utilities", "Clothing", "Transportation"]
 
-    classify = guidance(f"""
+    # define the prompt
+    classify = guidance("""
             {{#system~}}
             You are an expense classifier that responds back in valid JSON. \n
             Classify the expense, given the description, vendor name, and price. Include a \
@@ -31,8 +39,10 @@ async def guidance_classify(expense_input: ExpenseInput):
                 "name": "{{gen 'name'}}",
                 "category": "{{select 'category' options=valid_categories}}",
                 "justification": "{{gen 'justification'}}"
-            }```""", llm=classify_llm)
+            }```""")
 
-    output = classify(expense_input=expense_input)
+    # execute the prompt
+    output = classify(expense_input=expense_input, valid_categories=valid_categories)
     logging.info('Classified %s', output)
+
     return output

python/classify-expenses/settings.py

@@ -5,16 +5,14 @@
 
 
 class Settings(BaseSettings):
-    memory_collection_name = os.getenv("EMBEDDING_COLLECTION_NAME")
-    memory_embedding_model = os.getenv("OPENAI_EMBEDDING_MODEL")
-    memory_embedding_dimension = 1536
+    # memory_collection_name = os.getenv("EMBEDDING_COLLECTION_NAME")
+    # memory_embedding_model = os.getenv("OPENAI_EMBEDDING_MODEL")
+    # memory_embedding_dimension = 1536
 
     openai_api_type = "storage"
     openai_api_version = "2022-12-01"
     openai_api_base = os.getenv("OPENAI_API_BASE")
     openai_api_key = os.getenv("OPENAI_API_KEY")
 
 
-
-
 settings = Settings()

python/classify-expenses/setup.py

@@ -11,7 +11,7 @@
         "fastapi>=0.95.1",
         "pydantic>=1.10.7",
         "asyncpraw>=7.7.0",
-        "guidance>=0.0.57"
+        "guidance>=0.0.57",
         "memory-client>=1.1.7",
         "semantic-kernel>=0.2.7.dev0"
     ],

typescript/frontend/app/chat-session/route.ts

@@ -8,7 +8,10 @@ export async function GET(request: Request) {
     console.dir(userId)
     const res = await fetch(RECCOMMENDATION_SERVICE_URL, {
         method: 'POST',
-        headers: { 'Content-type': `application/json` }
+        headers: {
+            'Content-type': `application/json`,
+            'x-sk-api-key': `${process.env.NEXT_PUBLIC_SK_API_KEY}`
+        }
     });
 
     const data = await res.json();

typescript/frontend/app/chatSession/getAllChats/route.ts

@@ -8,7 +8,10 @@ export async function GET(request: Request) {
     console.dir(userId)
     const res = await fetch(RECCOMMENDATION_SERVICE_URL, {
         method: 'POST',
-        headers: { 'Content-type': `application/json` }
+        headers: {
+            'Content-type': `application/json`,
+            'x-sk-api-key': `${process.env.NEXT_PUBLIC_SK_API_KEY}`
+        }
     });
 
     const data = await res.json();

typescript/frontend/src/components/chat/chat-session-list.tsx

@@ -16,7 +16,15 @@ export function ChatSessionList({ className, setSelectedSession: updateSelectedS
 
     useEffect(() => {
         async function fetchChatSessions() {
-            const response = await fetch(`${process.env.NEXT_PUBLIC_COPILOT_CHAT_BASE_URL}/chatSession/getAllChats/${userInfo.userName}`);
+            const chatSessionUrl = `${process.env.NEXT_PUBLIC_COPILOT_CHAT_BASE_URL}/chatSession/getAllChats/${userInfo.userName}`;
+            console.log(chatSessionUrl);
+            const response = await fetch(chatSessionUrl, {
+                method: 'GET',
+                headers: {
+                    'Content-type': `application/json`,
+                    'x-sk-api-key': `${process.env.NEXT_PUBLIC_SK_API_KEY}`
+                }
+            });
             const data = await response.json();
             setChatSessions(data);
             console.log("Chat sessions");
@@ -27,7 +35,14 @@ export function ChatSessionList({ className, setSelectedSession: updateSelectedS
     }, []);
 
     async function fetchChatMessages(chatId: string) {
-        const response = await fetch(`${process.env.NEXT_PUBLIC_COPILOT_CHAT_BASE_URL}/chatSession/getChatMessages/${chatId}?startIdx=0&count=-1`);
+        const chatMsgEndpoint = `${process.env.NEXT_PUBLIC_COPILOT_CHAT_BASE_URL}/chatSession/getChatMessages/${chatId}?startIdx=0&count=-1`
+        const response = await fetch(chatMsgEndpoint, {
+            method: 'GET',
+            headers: {
+                'Content-type': `application/json`,
+                'x-sk-api-key': `${process.env.NEXT_PUBLIC_SK_API_KEY}`
+            }
+        });
         const data = await response.json();
         setChatsAtom(data as ChatProps[]);
     }

typescript/frontend/src/components/transaction/transaction-table.tsx

@@ -17,14 +17,14 @@ const COLUMNS = [
     maxWidth: 180,
   },
   {
-    Header: () => <div className="ltr:ml-auto rtl:mr-auto">GPT Category</div>,
+    Header: () => <div className="ltr:ml-auto rtl:mr-auto">Category (AI Generated)</div>,
     accessor: 'category',
     // @ts-ignore
     Cell: ({ cell: { value } }) => (
       <div className="ltr:text-right rtl:text-left">{value}</div>
     ),
-    minWidth: 100,
-    maxWidth: 220,
+    minWidth: 150,
+    maxWidth: 290,
   },
   {
     Header: () => <div className="ltr:ml-auto rtl:mr-auto">Description</div>,

typescript/frontend/src/layouts/sidebar/chat-blade.tsx

@@ -82,7 +82,8 @@ export default function Sidebar({ className, setSelectedSession, setUserInfoAtom
         const response = await fetch(`${process.env.NEXT_PUBLIC_COPILOT_CHAT_BASE_URL}/skills/ChatSkill/functions/Chat/invoke`, {
             method: 'POST',
             headers: {
-                'Content-Type': 'application/json',
+                'Content-type': `application/json`,
+                'x-sk-api-key': `${process.env.NEXT_PUBLIC_SK_API_KEY}`
             },
             body: JSON.stringify({
                 input: userInput,